Graph Neural Networks(GNNs),owing to their exceptional capability to capture high-order connectivity,have emerged as a leading technology in the domain of collaborative filtering.Despite their remarkable performance in modeling user-item interactions,current studies often neglect the homophily relationships between users and items,which are pivotal for en-hancing the performance of recommendation systems.Although GNNs can indirectly acquire homophily information through even-order neighbor nodes,the approach has its limitations,potentially leading to suboptimal recommendation outcomes and diffi-culties in accurately capturing homophily.To address these issues,we introduce the Homophily Relationship Enhanced Graph Con-volutional Network(HREGCN),which constructs a homophily graph via Singular Value Decomposition(SVD)to more precisely identify the homophily between users and items.Experimental results on two public datasets indicate that HREGCN has achieved significant performance improvements in graph-based collaborative filtering tasks.
万方数据